Multi-Project Risk Phrase Configuration Guide: Achieving Independent Internal Control Isolation for Each Telegram Project

When your team operates multiple Telegram bot projects simultaneously—for example, one for cross-border e-commerce customer service, one for Web3 community management, and another for internal ticket notifications—you’ll quickly find that a unified content moderation rule simply doesn’t work. Sensitive words for Project A (such as competitor brand names) may be meaningless for Project B; wallet addresses that Project C needs to monitor could inadvertently interfere with normal business communication in Project D.

That’s why TG-Staff Pro offers the Project Risk Phrases feature: you can create independent keyword libraries for each Telegram project, achieving precise internal control isolation. This guide will walk you through the entire process from creation to verification.

Why Do Multi-Project Environments Need Independent Risk Phrases?

Consider a real scenario: your team manages three bot projects—

• Project A: A cryptocurrency exchange customer service bot, needing to monitor if agents send personal wallet addresses.
• Project B: An e-commerce brand after-sales bot, needing to block competitor brand names (e.g., “XX same style cheaper”).
• Project C: An internal O&M notification bot, which hardly requires content moderation.

If all projects share the same set of risk words, what happens? When an agent in Project C sends “Please check the TRC20 transfer record,” the system might trigger a pop-up due to wallet address keywords, causing unnecessary disruption. Conversely, agents in Project A might miss detecting new scam addresses because the phrase list doesn’t include them.

The core value of Project Risk Phrases is solving this: each project independently associates a set of keywords without interfering with others. This not only improves agent efficiency but also ensures precise compliance and internal control—especially for Web3/cryptocurrency teams, where wallet address monitoring must be isolated by project to avoid false positives.

Step 1: Log In to TG-Staff Console and Access the Content Moderation Module

Check Your Plan and Permissions

First, confirm your TG-Staff plan is Pro. Content moderation (Internal Control Management) is a Pro-exclusive feature and is not included in the Standard plan.

• Log in to app.tg-staff.com.
• Click your avatar in the top right → Plan & Billing to view your current plan.
• If it shows “Standard,” click “Upgrade Plan” and select Pro (annual payment offers a discount; see the official plan page for details).

Locating the “Risk Phrases” Management Interface

After a successful upgrade, find Settings → Content Risk Control → Risk Phrases in the left navigation bar. This is the management center for all phrases.

• The page displays a list of all created phrases by default (empty on first use).
• The + New Phrase button is in the top right corner.
• Each phrase card shows: phrase name, number of associated projects, keyword count, and last modification time.

Step 2: Create Independent Risk Phrases for Each Project

Naming Conventions and Grouping Strategies

Click + New Phrase to enter the creation page. Naming conventions directly affect subsequent management efficiency; it is recommended to follow a unified format:

[项目名]-[用途]-[版本]

Example:

Phrase Name	Associated Project	Usage Description
ProjectA-WalletAddress-v1	Crypto Exchange Bot	Monitor TRC20/ERC20 wallet addresses
ProjectA-CompetitorBrands	Crypto Exchange Bot	Block competitor exchange names
ProjectB-AfterSalesSensitive	E-commerce Customer Service Bot	Block complaint escalation keywords
ProjectC-Disabled	Internal Ops Bot	Empty phrase (risk control not actually enabled)

With this naming, you can instantly identify in audit logs which project triggered a record and what type of risk it belongs to.

Adding Keywords: From Words to Address Fragments

After creating a phrase, enter the keyword addition interface. Two matching modes are supported:

• Exact word match: Enter “fraud” — triggers only when the message contains the exact word “fraud”.
• Fragment match: Enter “TXYZ123” — triggers when the message contains “TXYZ123” anywhere (e.g., “address is TXYZ12345”).

For wallet address monitoring scenarios, it is recommended to use address fragments instead of full addresses. For example, if you want to monitor a known fraudulent address TXYZ123456789abcdef, you can add the fragment TXYZ123456. This way, even if scammers use similar addresses (like TXYZ123456789abcde0), they will be blocked.

When adding keywords, enter one word per line. Supports Chinese, English, numbers, and symbols. Click Save after adding.

Step 3: Associate Risk Phrases with Specific Projects

Single Project Association Demo

Go back to the project list and enter the Bot project you need to configure (e.g., “Cryptocurrency Exchange Bot”):

1. Click the project name → Edit Project.
2. Find the Content Risk Control tab.
3. In the Risk Phrases dropdown, select the phrase you just created (e.g., “Project A - Wallet Address - v1”).
4. Click Save Project Settings.

Done! Now all outbound messages from agents in this project will be checked against this phrase.

Multi-Project Isolation Implementation

Assume you have two projects configured as follows:

Project	Associated Risk Phrase	Expected Behavior
Project A: Cryptocurrency Exchange Bot	Project A - Wallet Address - v1	Sending “Please transfer to TXYZ123” → triggers interception
Project B: E-commerce Customer Service Bot	Project B - After-Sales Sensitive Words	Sending “Please transfer to TXYZ123” → sent normally (no interception)

Isolation Effect: The same sentence is intercepted in Project A but sent normally in Project B. This demonstrates the power of independent risk phrases.

Note: A project can only be associated with one risk phrase. If you need multiple rule sets (e.g., monitoring both wallet addresses and competitor keywords), it is recommended to merge all keywords into the same phrase.

Step 4: Test and Verify Isolation Effect

After configuration, be sure to test. It is recommended to use two test agent accounts, each logging into different projects.

Specific operations:

1. Simulate trigger: In Project A’s chat interface, an agent sends a message containing a wallet address fragment and clicks send. The system should display a secondary confirmation popup (or directly block, depending on your trigger settings).
2. Cross-test: The same agent switches to Project B (if they have permission) and sends the same message → it should send normally.
3. Audit log: Go to Content Moderation → Trigger Records, and check if the log only shows Project A’s trigger records, including the trigger time, agent, risk word, etc.

If testing reveals that Project B also triggers interception, check: Is Project B incorrectly associated with Phrase Group A? Or does Phrase Group A contain overly generic fragments (e.g., “123”)?

Advanced Tips: Dynamic Adjustment and Audit Monitoring

Content moderation is not a “configure once, valid forever” solution. As your business evolves, you need to make dynamic adjustments:

• Add keywords: When new scam addresses or sensitive words are discovered, immediately add them to the corresponding phrase group.
• Remove false positives: If a keyword frequently causes false positives in normal business, remove it promptly or adjust it to a more precise fragment.
• Audit log analysis: Regularly review trigger records to analyze which agents trigger frequently and which risk words are hit most often. This helps optimize phrase group configuration and even uncover potential internal control vulnerabilities.

For example, audit logs show that “Project A - Wallet Address v1” was triggered 20 times in one day, with 18 times from the same agent. Further analysis reveals that the agent habitually sends personal wallet addresses when guiding users to transfer funds. You can then: ① Demote the agent’s permissions; ② Add more wallet address fragments commonly used by that agent to the phrase group; ③ Arrange compliance training.

Frequently Asked Questions

Q: I created a risk phrase group, but when an agent sends a trigger word, there is no response. What could be the reasons?

A: Check three points: ① Is the agent part of the project associated with the risk phrase group? ② Does the phrase group contain the trigger word (note case sensitivity and fragment matching)? ③ Is it within the Pro plan (Standard plan does not support this feature)?

Q: Can a project be associated with multiple risk phrase groups?

A: Currently, each project supports only one risk phrase group. If you need multiple sets of rules, consider merging all keywords into one phrase group or splitting projects based on needs. For example, you can create a “Project A - Comprehensive Moderation” phrase group containing both wallet addresses and competitor keywords.

Q: Do modifications to risk phrase groups take effect immediately?

A: Yes, they take effect immediately after saving, without needing to restart the project or log in again. However, it is recommended to make adjustments during off-peak hours to avoid affecting agent experience.

Q: How can I batch import a large number of risk words?

A: Currently, manual entry one by one is supported. For batch import, please contact @tgstaff_robot to provide feedback; CSV import may be supported in the future.

Q: Do risk phrase groups support regular expressions?

A: The current version only supports keyword fragment matching (including full words and fragments). Regular expressions are not supported. If fuzzy matching is needed, add multiple variants. For example, to match “scam”, “fraud”, and “deception”, add these three words separately.

---

Experience multi-project independent internal controls now: Sign up for TG-Staff Pro and configure exclusive risk phrase groups for each of your Telegram projects.
👉 https://app.tg-staff.com/
📖 View full documentation
💬 Have questions? Contact @tgstaff_robot